Strategy tuning of synthetic jet control over supercritical airfoil based on deep reinforcement learning

In this article, the author couples Soft Actor-Critic (SAC), a deep reinforcement learning (DRL) algorithm, and synthetic jet (SJ) control, to explore the application of SAC algorithm in the field of active flow control (AFC). The author develops a framework of coupling DRL algorithms and computational fluid dynamics (CFD), in which a novel DRL environment is build based on ANSYS Fluent, and interacts with DRL agent using PyFluent kit as a bridge. In the case of two-dimensional (2D) cylinder laminar flow with Reynolds number (Re) = 100, SAC algorithm achieves 33.3% improvement of drag reduction compared with Proximal Policy Optimization (PPO) algorithm, another DRL algorithm, with 70% of training time consumption. The newly developed framework is adopted in tuning the strategy of SJ control over super critical airfoil. With the objective to enhance lift, agent's strategy gains 3.15% extra lift compared with the baseline which is not stimulated by SJ, and 29.03% improvement of lift enhancement compared with stable SJ frequency strategy. In another case involving a double-objective optimization, where both lift enhancement and drag reduction are considered as objectives, agent's strategy achieves 49.088 of time-averaged lift-drag ratio, which is 10.31% more than that of the baseline, and 1.58% – 14.7% improvement compared with stable SJ frequency strategy. With the foundation of the DRL environment introduced in this article, researchers can easily migrate other DRL algorithms into this framework and do further application in AFC.